Model-free adaptive temperature control method and system in infrared heating process

Abstract

The invention relates to a model-free adaptive temperature control method and system in an infrared heating process, and relates to the technical field of infrared heating control regulation, and the method comprises the steps: describing an infrared heating system as a nonlinear system of discrete time, and constructing a corresponding linearization data model; and then introducing an RBF neural network to estimate the dynamic characteristics of the infrared heating system, and estimating control law parameters on line. A rolling optimization prediction mechanism of model prediction control is used to process time lag characteristics in an infrared heating process, and an RBF-MFAPC controller is constructed. And finally, taking the data acquired in real time as the input of the controller, and dynamically adjusting the power of the infrared heater through real-time feedback, thereby realizing the accurate control of the temperature in the infrared heating process. The method does not need to depend on model parameters in the infrared heating process, achieves the high-precision control of the temperature in the infrared heating process through the nonlinear approximation system dynamic characteristics of the RBF neural network and a rolling optimization mechanism of model prediction control, and improves the adaptability of the system.

Description

Technical Field

[0001] The present application relates to the field of infrared heating control and regulation technology, and in particular to a model-free adaptive temperature control method and system for an infrared heating process. Background Art

[0002] Among related technologies, infrared heating technology utilizes infrared radiation to directly heat objects, eliminating the intermediate energy transfer required in traditional heating methods. This significantly improves energy efficiency and reduces energy waste. Its clean, pollution-free nature also makes it an ideal heating method that meets environmental requirements. With its unique high efficiency, energy-saving, and environmentally friendly features, infrared heating technology has gradually become a key pillar of modern manufacturing and green technology.

[0003] Infrared heating systems exhibit significant nonlinearity and model uncertainty in industrial applications, making it difficult for traditional linear control methods and model-based temperature control methods to achieve ideal control effects. First, the infrared heating process involves the complex coupling of thermal radiation fields and temperature fields, and the relationship between heating power and temperature exhibits significant nonlinear characteristics. This nonlinear characteristic makes it difficult to accurately describe and regulate the dynamic changes of the system using traditional linear control methods. Second, during the heating process of the infrared heating system, the thermophysical properties of the material (such as thermal conductivity and specific heat capacity) will change significantly with time and temperature, making it difficult to obtain the precise model parameters of the system and causing uncertainty.

[0004] To address these issues, existing control methods for infrared heating systems primarily transform the infrared heating process into an approximate mathematical model to achieve model-based control. However, these models often fail to fully reflect the dynamic characteristics of the actual system, resulting in limited effectiveness of model-based control methods in practical applications. Consequently, existing technologies often exhibit problems in infrared heating systems, such as poor controllability, weak adaptability, and low precision. Summary of the Invention

[0005] The present application provides a model-free adaptive temperature control method and system for an infrared heating process, which addresses the problems of poor controllability, weak adaptability, and low precision in existing system model-based control strategies in infrared heating systems. The present application adopts a model-free adaptive control strategy (RBF-MFAPC) based on RBF neural network and model predictive control, which achieves no reliance on model parameters of the infrared heating process. Through the nonlinear approximation of the system dynamic characteristics of the RBF neural network and the rolling optimization mechanism of the model predictive control, the control accuracy of the MFAC is significantly improved, thereby achieving high-precision temperature control during the infrared heating process and improving the adaptability of the system.

[0006] In a first aspect, the present application provides a model-free adaptive temperature control method for an infrared heating process, comprising:

[0007] According to input and output data of the infrared heating system, the infrared heating system is converted into a nonlinear discrete time system, and the nonlinear discrete time system is converted into a dynamic linear data model;

[0008] Using an RBF neural network to perform online estimation of pseudo partial derivatives in the dynamic linear data model to obtain estimated pseudo partial derivative values, wherein the pseudo partial derivative values ​​are output parameters obtained by the RBF neural network using nonlinear approximation capabilities to estimate the dynamic characteristics of the infrared heating system;

[0009] Using MFAC-based model predictive control to process the time lag characteristics of the infrared heating system during the infrared heating process, and constructing a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data;

[0010] Taking the real-time collected heating data of the infrared heating system as input, the predictive controller analyzes and predicts the behavior of the infrared heating system within a certain period of time, optimizes the control input, and outputs dynamic adjustment information;

[0011] The dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

[0012] Optionally, according to input and output data of the infrared heating system, converting the infrared heating system into a nonlinear discrete time system, and converting the nonlinear discrete time system into a dynamic linear data model, comprises:

[0013] Based on the input and output data of the selected infrared heating system,

[0014] y(k+1)=f(y(k),y(k-1),...y(kn y );u(k),u(k-1),...u(kn u )), converting the infrared heating system into a nonlinear discrete-time system;

[0015] According to y(k+1)=y(k)+φ(k)Δu(k), the nonlinear discrete system of infrared heating process is converted into a compact dynamic linear data model;

[0016] Where k is the sampling time, y(k+1) is the actual output temperature of the infrared heating system at the next moment, f is an unknown nonlinear function, y(k) is the output temperature of the infrared heating system at the current moment, u(k) is the input power, y(k-1) is the output temperature of the infrared heating system at the previous moment, u(k-1) is the input power of the infrared heating system at the next moment, and n y Indicates the order of the system output, n u represents the order of the system input, φ(k) is the pseudo partial derivative of the system, Δu(k)=u(k)-u(k-1) and Δu(k)≠0.

[0017] Optionally, an RBF neural network is used to perform online estimation of the pseudo partial derivative in the dynamic linear data model to obtain an estimated pseudo partial derivative value, including:

[0018] The information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between the neurons to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative value estimated at the current moment;

[0019] In the process of estimating the pseudo partial derivative, the one-step-ahead square error is used as a performance indicator function to feed back and update the connection weights between the hidden layer neurons and the output layer neurons at the next moment.

[0020] Optionally, the information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between the neurons to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative value estimated at the current moment, including:

[0021] According to x(k)=[y(k),…,y(km);u(k-1),…,u(kn)], construct the input and output data information vector as the RBF neural network input x(k);

[0022] The Gaussian basis function is used as the activation function, according to The input x(k) is nonlinearly mapped and The output of the hidden layer neurons is linearly superimposed to obtain the output layer output φ(k) of the RBF neural network as the estimated pseudo partial derivative;

[0023] Among them, φ(k) is the estimated pseudo partial derivative in the data model, m and n are both positive integers, representing the number of input and output data, c j represents the center vector of the jth hidden layer neuron; bj Indicates the radius of the jth hidden layer neuron, ω j (k) represents the connection weight between the jth neuron in the hidden layer and the neuron in the output layer at the current moment.

[0024] Optionally, a model predictive control based on MFAC is used to process the time lag characteristics of the infrared heating system during the infrared heating process, and a model-free adaptive predictive controller is constructed based on the pseudo partial derivative and the input and output data, including:

[0025] Taking the pseudo partial derivative and the real-time collected temperature data as input, according to Define a model-free adaptive predictive controller;

[0026] according to Define the N-step forward prediction output equation of the infrared heating system;

[0027] Where N represents the output prediction step size of the system, represents the predicted output matrix of the system, represents the expected output matrix of the system, N u represents the control time domain constant of the system; Represents the predicted input change matrix of the system, E(k) represents the identity matrix, and A(k) represents the pseudo partial derivative prediction matrix.

[0028] Optionally, the infrared heating system's heating data collected in real time is used as input, the prediction controller analyzes and predicts the infrared heating system's behavior within a certain time period, optimizes the control input, and outputs dynamic adjustment information, including:

[0029] Modeling the time series variation law of the pseudo partial derivative value using a pseudo partial derivative prediction algorithm, updating the pseudo partial derivative prediction value in combination with prediction error feedback correction, and updating the prediction matrix in the prediction controller using the pseudo partial derivative prediction value;

[0030] The dynamic linear data model, the updated pseudo partial derivative prediction matrix and the real-time collected heating data are substituted into the control input criterion function in the predictive controller to construct a control law expression, and the first control quantity in the predicted input change matrix is ​​used to update the predictive control strategy and output dynamic adjustment information.

[0031] Optionally, a pseudo partial derivative prediction algorithm is used to model the time series variation law of the pseudo partial derivative value, the pseudo partial derivative prediction value is updated in combination with prediction error feedback correction, and the pseudo partial derivative prediction value is used to update the prediction matrix in the prediction controller, including:

[0032] according to Define the autoregressive model parameter matrix θ(k) and the autoregressive model pseudo partial derivative matrix Φ(k-1);

[0033] according to Get the autoregressive model parameters θ;

[0034] Taking the autoregressive model parameter θ as input, according to

[0035] φ(k+j)=θ1(k)φ(k+j-1)+θ2(k)φ(k+j-2)+…+θ p (k)φ(k+jp) constructs a pseudo partial derivative prediction algorithm and obtains the pseudo partial derivative prediction value φ(k+j);

[0036] Where δ is a positive number and δ∈(0,1], φ(k+j) represents the pseudo partial derivative prediction value at time j, p represents the order of the autoregressive model, and j=1.

[0037] Optionally, the dynamic linear data model, the updated pseudo partial derivative prediction matrix, and the real-time collected heating data are substituted into the control input criterion function in the predictive controller to construct a control law expression, and the first control variable in the predicted input change matrix is ​​used to update the predictive control strategy and output dynamic adjustment information, including:

[0038] according to

[0039] Construct control input criterion function J u ;

[0040] Substitute the dynamic linear data model and the real-time collected heating data into the control input criterion function J u , and predict the input change matrix Derivative, construct control law expression

[0041]

[0042] by For input, according to Predict the first control variable in the input change matrix and obtain the control input u(k) at the current moment as dynamic adjustment information;

[0043] Among them, λ is the penalty factor of the system input, which is used to limit the variable of the control input and ensure the smoothness of the control input.

[0044] Optionally, after outputting the dynamic adjustment information, the following is also included:

[0045] The one-step-ahead square error is used as the performance indicator function J, according to Determine the expected output signal of the system at the next moment;

[0046] Minimize the performance index function J, according to Updating the connection weights, wherein the updated connection weights are used to update the pseudo partial derivative values;

[0047] Among them, y(k+1) is the actual output signal of the system at the next moment, and the connection weight ω from the hidden layer to the output layer is corrected by the gradient descent method. j ,ω j (k+1) represents the connection weight at the next moment; α is the learning efficiency, and α∈(0,1], β is the inertia coefficient, and β∈(0,1], ω j (k-1) represents the connection weight at the previous moment, Δω j (k)=ω j (k)-ω j (k-1).

[0048] In a second aspect, the present application provides a model-free adaptive temperature control system for an infrared heating process, comprising:

[0049] a conversion module, configured to convert the infrared heating system into a nonlinear discrete time system according to input and output data of the infrared heating system, and to convert the nonlinear discrete time system into a dynamic linear data model;

[0050] a pseudo partial derivative estimation module, configured to perform online estimation of the pseudo partial derivatives in the dynamic linear data model using an RBF neural network to obtain estimated pseudo partial derivative values, wherein the pseudo partial derivative values ​​are output parameters obtained by the RBF neural network using a nonlinear approximation capability to estimate the dynamic characteristics of the infrared heating system;

[0051] A predictive controller construction module is used to process the time lag characteristics of the infrared heating system during the infrared heating process by using the MFAC-based model predictive control, and to construct a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data;

[0052] an analysis and prediction module, configured to use the real-time collected heating data of the infrared heating system as input, analyze and predict the behavior of the infrared heating system within a certain time period through the prediction controller, optimize the control input, and output dynamic adjustment information;

[0053] The dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

[0054] In summary, the embodiment of the present application first describes the infrared heating system as a discrete-time nonlinear system, and constructs a linearized data model of the infrared heating system through a dynamic linearization method. Then, an RBF neural network is introduced to online estimate the dynamic characteristics of the infrared heating system, and online estimation of the control law parameters is realized. The rolling optimization prediction mechanism of the model predictive control (MPC) is used to process the time-lag characteristics in the infrared heating process, and the design of the RBF-MFAPC controller is completed. Finally, the real-time collected temperature signal is used as the input of the RBF-MFAPC controller, and the infrared heater power is dynamically adjusted through real-time feedback, thereby achieving precise control of the temperature of the infrared heating process. This embodiment does not need to rely on the model parameters of the infrared heating process. Through the nonlinear approximation of the system dynamic characteristics of the RBF neural network and the rolling optimization mechanism of the model predictive control, the control accuracy of the MFAC is significantly improved, high-precision temperature control of the infrared heating process is achieved, and the adaptability of the system is improved, solving the problems of poor controllability, weak adaptability and low accuracy in the infrared heating system of the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0056] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0057] Figure 1 A schematic flow chart of a model-free adaptive temperature control method for an infrared heating process provided in an embodiment of the present application;

[0058] Figure 2 This is a schematic flow chart of the steps of a model-free adaptive temperature control method for an infrared heating process provided by an optional embodiment of the present application;

[0059] Figure 3 This is a BRF-MFAPC control block diagram provided by an optional example of this application;

[0060] Figure 4 This is a structural block diagram of a model-free adaptive temperature control system for an infrared heating process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0061] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0062] To facilitate understanding of the embodiments of the present application, further explanation will be given below in conjunction with the drawings and specific embodiments. The embodiments do not constitute a limitation on the embodiments of the present application.

[0063] Figure 1 The following is a flow chart of a model-free adaptive temperature control method for an infrared heating process provided in an embodiment of the present application. Figure 1 As shown, the model-free adaptive temperature control method for the infrared heating process provided in the embodiment of the present application may specifically include the following steps:

[0064] Step 110 : converting the infrared heating system into a nonlinear discrete time system according to input and output data of the infrared heating system, and converting the nonlinear discrete time system into a dynamic linear data model.

[0065] In this embodiment, the input and output data of the infrared heating system include but are not limited to: temperature, power data and power instructions (used to control the power of the infrared heating system and thus the heating temperature) during the infrared heating process.

[0066] In a specific implementation, this embodiment describes the heating process of the infrared heating system as a discrete-time nonlinear system (or nonlinear discrete-time system) based on the input and output data of the infrared heating system. Considering that the partial derivatives of the system input variables are continuous, this embodiment further transforms the nonlinear discrete system of the infrared heating process into a compact dynamic linearized data model (referred to as the data model).

[0067] Step 120: Use an RBF neural network to perform online estimation on the pseudo partial derivatives in the dynamic linear data model to obtain estimated pseudo partial derivative values.

[0068] The pseudo partial derivative is an output parameter obtained by estimating the dynamic characteristics of the infrared heating system using the nonlinear approximation capability of the RBF neural network.

[0069] In a specific implementation, in order to solve the problems existing in the existing infrared system heating control strategy based on the system model, this embodiment adopts Model Free Adaptive Control (MFAC) to realize adaptive temperature control.

[0070] Because the default pseudo-partial derivative parameter estimation algorithm in the original MFAC is a projection algorithm, when the controlled system is in a complex operating environment, the traditional projection algorithm relies on a fixed step size or empirical parameters to dynamically update the pseudo-partial derivative values. In scenarios with strong nonlinearities and multiple disturbance coupling, such as infrared heating, sudden changes in dynamic characteristics can easily lead to accumulated estimation errors, which can cause control instability (such as temperature overshoot or oscillation).

[0071] Therefore, this embodiment introduces an RBF neural network to estimate the dynamic characteristics of infrared heating systems. Specifically, RBF neural networks, leveraging the nonlinear mapping properties of the Gaussian kernel function in the hidden layer and the linear weighting properties of the output layer, can effectively approximate the dynamic behavior of complex systems. Therefore, this embodiment embeds the RBF neural network into the MFAC framework, replacing the traditional projection algorithm used in existing technologies.

[0072] In specific implementations, RBF neural networks can utilize their excellent nonlinear approximation capabilities to accurately estimate the system's dynamics, thereby improving the system's control accuracy. Specifically, RBF neural networks can use the system's input and output data to perform linear superposition on the input data using the hidden layer neurons in the RBF neural network, thereby estimating the pseudo-partial derivatives in the dynamic linear data model, obtaining pseudo-partial derivative values, and automatically adjusting and optimizing the pseudo-partial derivative values ​​estimated by the system to adapt to more diverse and complex working conditions, thereby improving the adaptability and control accuracy of infrared heating systems in complex working environments.

[0073] Step 130 : Using MFAC-based model predictive control to process the time lag characteristics of the infrared heating system during the infrared heating process, and constructing a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data.

[0074] In order to better deal with the time lag problem in the infrared heating process, this embodiment first integrates the rolling optimization mechanism of model predictive control (MPC), which is commonly used to solve time lag, into the control framework of MFAC. Then, the RBF neural network and the prediction mechanism of model predictive control are introduced into the MFAC framework to implement the prediction algorithm based on RBF-MPAPC.

[0075] This embodiment uses the nonlinear approximation capability of the RBF neural network to accurately estimate the dynamic characteristics of the system, and integrates the advantage of model predictive control in predicting the future behavior of the system, providing an effective time-delay response strategy for the infrared heating process, thereby achieving high-precision temperature control of the infrared heating process.

[0076] Step 140 , using the real-time collected heating data of the infrared heating system as input, the prediction controller analyzes and predicts the behavior of the infrared heating system within a certain period of time, optimizes the control input, and outputs dynamic adjustment information.

[0077] The dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

[0078] In its specific implementation, the RBF-MFAPC prediction algorithm does not need to rely on an accurate mathematical model of the infrared heating process. It uses the real-time heating data of the infrared heating system during the infrared heating process as a benchmark to predict the system's behavior in the future and optimize the control input to compensate for the time lag effect. This includes predicting the changes in the system's output and input signals at N moments in the future, and outputting corresponding dynamic adjustment information to control the heating power of the infrared heating system during the infrared heating process. This achieves precise optimization of the control input at future moments without relying on an accurate model, has strong adaptability, and can effectively solve the time lag and model uncertainty problems of complex nonlinear systems.

[0079] It can be seen that the embodiment of the present application first describes the infrared heating system as a discrete-time nonlinear system, and constructs a linearized data model of the infrared heating system through a dynamic linearization method. Then, an RBF neural network is introduced to estimate the dynamic characteristics of the infrared heating system online, and online estimation of the control law parameters is realized. The rolling optimization prediction mechanism of the model predictive control (MPC) is used to process the time lag characteristics in the infrared heating process, and the design of the RBF-MFAPC controller is completed. Finally, the real-time collected temperature signal is used as the input of the RBF-MFAPC controller, and the infrared heater power is dynamically adjusted through real-time feedback, thereby achieving precise control of the temperature of the infrared heating process. This embodiment does not need to rely on the model parameters of the infrared heating process. Through the nonlinear approximation of the system dynamic characteristics of the RBF neural network and the rolling optimization mechanism of the model predictive control, the control accuracy of the MFAC is significantly improved, high-precision temperature control in the infrared heating process is achieved, and the adaptability of the system is improved, solving the problems of poor controllability, weak adaptability and low accuracy in the infrared heating system of the prior art.

[0080] Reference Figure 2 , shows a schematic flow chart of the steps of a model-free adaptive temperature control method for an infrared heating process provided by an optional embodiment of the present application. The method may specifically include the following steps:

[0081] Step 210 : converting the infrared heating system into a nonlinear discrete time system according to input and output data of the infrared heating system, and converting the nonlinear discrete time system into a dynamic linear data model.

[0082] In a specific implementation, this embodiment fully analyzes each infrared heating process of the infrared heating system, constructs a data model of the infrared heating process based on a compact dynamic linearization model, and describes the infrared heating process as a discrete-time nonlinear system.

[0083] In a specific implementation, in order to accurately represent the infrared heating process as a nonlinear discrete-time system, this embodiment fully considers the input and output of the infrared heating system at the current moment and the next moment during the infrared heating process, and defines the nonlinear discrete-time system as:

[0084] y(k+1)=f(y(k),y(k-1),...y(kn y );u(k),u(k-1),...u(kn u )).

[0085] Optionally, the embodiment of the present application converts the infrared heating system into a nonlinear discrete time system according to the input and output data of the infrared heating system, and converts the nonlinear discrete time system into a dynamic linear data model, which may include: taking the input and output data of the selected infrared heating system as a reference, according to y(k+1)=f(y(k),y(k-1),...y(kn y );u(k),u(k-1),...u(kn u )), convert the infrared heating system into a nonlinear discrete-time system; according to y(k+1)=y(k)+φ(k)Δu(k), convert the nonlinear discrete system of the infrared heating process into a compact dynamic linear data model; wherein k is the sampling time, y(k+1) is the actual output temperature of the infrared heating system at the next moment, f is an unknown nonlinear function, y(k) is the output temperature of the infrared heating system at the current moment, u(k) is the input power, y(k-1) is the output temperature of the infrared heating system at the previous moment, u(k-1) is the input power of the infrared heating system at the next moment, and n y Indicates the order of the system output, n u represents the order of the system input, φ(k) is the pseudo partial derivative of the system, Δu(k)=u(k)-u(k-1) and Δu(k)≠0.

[0086] In actual implementation, considering that the partial derivatives of the nonlinear function f with respect to the system input variables are continuous and satisfy the generalized Lipschitz condition, the nonlinear discrete system of the infrared heating process in the above formula can be converted into a compact dynamic linearization data model, so that the pseudo-partial derivatives of the system can be accurately obtained, where the pseudo-partial derivatives must be less than a certain positive constant.

[0087] In step 220, the information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between the neurons to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative value estimated at the current moment.

[0088] In related technologies, model-free adaptive control (MFAC) algorithms can meet the control needs of nonlinear systems with unknown models. This is because MFAC is not restricted by the mathematical model of the controlled object. Instead, it directly designs the controller based on the input and output data of the controlled object and performs parallel adaptive adjustments, effectively addressing model uncertainty and nonlinearity.

[0089] However, existing model-free adaptive control algorithms primarily rely on online estimation of pseudo-partial derivatives using a projection algorithm to achieve adaptive control of the system. However, the heating process of infrared heating systems is complex, and under these conditions, the estimation of pseudo-partial derivatives can be affected by external interference and become inaccurate. Pseudo-partial derivatives reflect the dynamic relationship between the system input and output, and their estimation accuracy directly determines the performance of the control algorithm. Therefore, when the pseudo-partial derivatives are inaccurately estimated, the controller's adjustment capability is limited, resulting in slow system response, increased overshoot, and even loss of control.

[0090] In response to the problems of inaccurate, untimely, and easy loss of control caused by the projection algorithm used in the existing technology, this embodiment proposes a control strategy that can accurately estimate pseudo-partial derivatives for complex working conditions. In addition, considering that in the infrared heating process, time lag problems such as temperature sensor detection lag and heat transfer delay are also important factors affecting the control performance of the system. This lag effect not only reduces the control accuracy of the system, but may also cause overshoot, oscillation, and even system instability. In order to overcome the adverse effects of time lag on the control performance of the infrared heating system, it is particularly necessary to propose a control algorithm that can effectively handle time lag.

[0091] Therefore, to address the problems caused by the projection algorithm used in existing technologies and to overcome the time lag in infrared heating system control, this embodiment embeds an RBF neural network into the MFAC framework, replacing the projection algorithm used in existing technologies. RBF neural networks utilize the nonlinear mapping properties of the hidden layer Gaussian basis functions and the linear weighting properties of the output layer to effectively approximate the dynamic behavior of complex systems.

[0092] Specifically, this embodiment uses the input and output data information vectors of the controlled system (i.e., the selected infrared heating system) as the input of the RBF neural network. Then, using the Gaussian basis function as the activation function, the output of the hidden layer neurons is determined through nonlinear mapping. The hidden layer is then processed and the output of the hidden layer neurons is linearly superimposed. During the linear superposition, the connection weights between the neurons are used to process the output of the neural network, i.e., the value of the estimated pseudo-partial derivative in the MFAC data model.

[0093] In an optional embodiment, the information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between the neurons to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative value estimated at the current moment, which may include: constructing the input and output data information vector according to x(k) = [y(k), ..., y(km); u(k-1), ..., u(kn)] as the input x(k) of the RBF neural network; using the Gaussian basis function as the activation function, according to The input x(k) is nonlinearly mapped and The output of the hidden layer neurons is linearly superimposed to obtain the output layer output φ(k) of the RBF neural network as the estimated pseudo partial derivative value; where φ(k) is the estimated pseudo partial derivative value in the data model, m and n are both positive integers, representing the number of input and output data, c j represents the center vector of the jth hidden layer neuron; b j Indicates the radius of the jth hidden layer neuron, ω j (k) represents the connection weight between the jth neuron in the hidden layer and the neuron in the output layer at the current moment.

[0094] In the specific implementation, in order to enable the RBF neural network to accurately estimate the dynamic characteristics of the system and improve the subsequent system control accuracy, this embodiment uses Accurately estimate the pseudo partial derivative value, and in order to improve the accuracy of the pseudo partial derivative estimation, this embodiment introduces the connection weight ω between the jth neuron in the hidden layer and the neuron in the output layer j (k), which is used to estimate the pseudo partial derivatives.

[0095] Therefore, this embodiment utilizes the excellent nonlinear approximation capability of the RBF neural network to accurately estimate the dynamic characteristics of the system, thereby improving the control accuracy of the system.

[0096] In step 230 , the MFAC-based model predictive control is used to process the time lag characteristics of the infrared heating system during the infrared heating process, and a model-free adaptive predictive controller is constructed based on the pseudo partial derivative and the input and output data.

[0097] In an optional embodiment, the MFAC-based model predictive control is used to process the time lag characteristics of the infrared heating system during the infrared heating process, and the pseudo partial derivative and the input and output data are used as a benchmark to construct a model-free adaptive predictive controller, including: taking the pseudo partial derivative and the real-time collected temperature data as input, according to Define a model-free adaptive predictive controller; according to Y N (k+1)=E(k)y(k)+A(k)ΔU Nu (k), define the N-step forward prediction output equation of the infrared heating system; where N represents the output prediction step size of the system, represents the predicted output matrix of the system, represents the expected output matrix of the system, N u represents the control time domain constant of the system; Represents the predicted input change matrix of the system, E(k) represents the identity matrix, and A(k) represents the pseudo partial derivative prediction matrix.

[0098] In the specific implementation, in order to better deal with the time lag problem in the infrared heating process, this embodiment integrates the rolling optimization mechanism of the model predictive control that solves the time lag problem into the control framework of MFAC, and compensates for the time lag effect by predicting the system behavior in the future and optimizing the control input.

[0099] In terms of specific implementation, in order to integrate the rolling optimization mechanism of model predictive control into the control framework of MFAC, this embodiment first defines the following system N (N is a preset value) step-forward output prediction equation based on the MFAC data model:

[0100]

[0101] In order to express the controller more concisely, it is converted into matrix form and a model-free adaptive predictive controller RBF-MFAC is constructed.

[0102] Step 240, using a pseudo partial derivative prediction algorithm to model the time series variation law of the pseudo partial derivative value, combining the prediction error feedback correction to update the pseudo partial derivative prediction value, and using the pseudo partial derivative prediction value to update the prediction matrix in the prediction controller.

[0103] In the specific implementation, this embodiment uses a predictive control strategy to optimize the control input at future times. In order to implement the optimization strategy, it is necessary to obtain the predicted value of the pseudo partial derivative. Therefore, this embodiment uses a pseudo partial derivative prediction algorithm to predict the pseudo partial derivative value of the system at future times, thereby updating the prediction matrix, where the prediction matrix can be a pseudo Jacobian matrix.

[0104] In actual implementation, the autoregressive model can be used to model the time series variation law of the pseudo partial derivative, combined with the prediction error feedback correction, to construct a prediction algorithm for the pseudo partial derivative, thereby achieving high-precision recursion.

[0105] In an optional embodiment, using a pseudo partial derivative prediction algorithm to model the time series variation law of the pseudo partial derivative value, combining the prediction error feedback correction to update the pseudo partial derivative prediction value, and using the pseudo partial derivative prediction value to update the prediction matrix in the prediction controller, can include: according to Define the autoregressive model parameter matrix θ(k) and the autoregressive model pseudo-partial derivative matrix Φ(k-1); according to Get the autoregressive model parameter θ; take the autoregressive model parameter θ as input, according to

[0106] φ(k+j)=θ1(k)φ(k+j-1)+θ2(k)φ(k+j-2)+…+θ p (k)φ(k+jp) constructs a pseudo partial derivative prediction algorithm to obtain the pseudo partial derivative prediction value φ(k+j); where δ is a positive number and δ∈(0,1], φ(k+j) represents the pseudo partial derivative prediction value at time j, p represents the order of the autoregressive model, and j=1.

[0107] In this embodiment, the pseudo partial derivative prediction value φ(k+j) is used to update the pseudo Jacobian matrix A(k).

[0108] Step 250, substitute the dynamic linear data model, the updated pseudo partial derivative prediction matrix and the real-time collected heating data into the control input criterion function in the predictive controller, construct the control law expression, and use the first control quantity in the predicted input change matrix to update the predictive control strategy and output dynamic adjustment information.

[0109] In the specific implementation, this embodiment substitutes the dynamic linear data model, the updated pseudo-partial derivative prediction matrix and the real-time collected heating data into the control input criterion function in the predictive controller to construct a control law expression. In order to enhance the control subject's ability to respond to environmental changes, only the first control quantity in the predicted input change matrix is ​​used to obtain the control input at the current moment. In the design process of the control law in this embodiment, the MFAC dynamic linearization data model is utilized, and the idea of ​​predictive control strategy to optimize the control input at future moments is combined. It does not rely on an accurate model, has strong adaptability, and can effectively solve the problems of time lag and model uncertainty of complex nonlinear systems.

[0110] Therefore, this embodiment realizes the introduction of the prediction mechanism of RBF neural network and model predictive control in the MFAC framework, accurately estimates the complex dynamic characteristics of the infrared heating process, constructs a dynamic linearized data model that adapts the system to the current state, and improves the control strategy's ability to cope with time-delay systems. The strategy decision can not only accurately estimate the dynamic characteristics of the system, but also compensate for the system's time delay, thereby overcoming the shortcomings of the traditional MFAC method, such as weak adaptability and inability to solve the time delay problem.

[0111] In an optional embodiment, the dynamic linear data model, the updated pseudo partial derivative prediction matrix, and the real-time collected heating data are substituted into the control input criterion function in the predictive controller to construct a control law expression, and the first control quantity in the predicted input change matrix is ​​used to update the predictive control strategy and output dynamic adjustment information, including: according to Construct control input criterion function J u ; Substitute the dynamic linear data model and the real-time collected heating data into the control input criterion function J u , and predict the input change matrix Derivative, construct control law expression by For input, according to Predict the first control variable in the input change matrix and obtain the control input u(k) at the current moment as dynamic adjustment information; where λ is the penalty factor of the system input, which is used to limit the variable of the control input and ensure the smoothness of the control input.

[0112]

[0113] Among them, A(k) is the pseudo-Jacobian matrix, which is updated by the pseudo partial derivative prediction value φ(k+j).

[0114] Among them, Take the derivative and set it equal to zero to obtain the corresponding control law expression.

[0115] Furthermore, in order to more accurately predict the output of the system at the next moment, after predicting the dynamic adjustment information, the connection weights from the hidden layer to the output layer can be updated, and the one-step-ahead square error can be used as the performance indicator function to feedback update the connection weights between the hidden layer neurons and the output layer neurons at the next moment to obtain the pseudo partial derivative at the next moment.

[0116] Specifically, to enable feedback updates of connection weights, this embodiment uses the one-step-ahead squared error as a performance indicator function and minimizes it. This minimization performance indicator function uses the system's output signal at the next moment (including the expected output signal and the actual output signal) as input and uses gradient descent to correct and update the connection weights.

[0117] Optionally, after outputting the dynamic adjustment information, the method may further include: taking the one-step-ahead square error as the performance indicator function J, and Determine the expected output signal of the system at the next moment; minimize the performance index function J, according to Update the connection weights, and the updated connection weights are used to update the pseudo partial derivative values; where y(k+1) is the actual output signal of the system at the next moment, and the connection weights ω from the hidden layer to the output layer are corrected by the gradient descent method. j ,ω j (k+1) represents the connection weight at the next moment; α is the learning efficiency, and α∈(0,1], β is the inertia coefficient, and β∈(0,1], ω j (k-1) represents the connection weight at the previous moment, Δω j (k)=ω j (k)-ω j (k-1).

[0118] To update the connection weights, this embodiment uses a minimization performance indicator function to construct an update algorithm formula for the connection weights:

[0119] For example, refer to Figure 3 As shown in the RBF-MFAPC control block diagram, in a specific implementation, the RBF-MFAPC algorithm of this embodiment first obtains input data when predicting the control strategy of the infrared heating system, including: controller parameters (b j 、c j , N, N u , p, λ, δ); temperature and power data information y(k),…,y(km); u(k-1),…,u(kn); expected signal y d; control the total time T. Then, the system pseudo partial derivative φ(k) is calculated using the RBF neural network pseudo partial derivative estimator, and the pseudo partial derivative prediction algorithm is used to predict the system pseudo partial derivative predicted value φ(k+j) at the subsequent time, thereby updating the parameter matrix θ(k) and pseudo Jacobian matrix A(k) of the autoregressive model. Finally, the power command u(k) is calculated using the model-free adaptive predictive controller, and the connection weight ω of the RBF neural network is updated. j .

[0120] In summary, the embodiment of the present application adopts a model-free adaptive control strategy (RBF-MFAPC) based on RBF neural network and model predictive control to address the nonlinearity, model uncertainty and time lag problems in the infrared heating system. First, the infrared heating process is described as a discrete-time nonlinear system and converted into a dynamic linearized data model of MFAC. Then, in order to more accurately estimate the dynamic characteristics of the system, the RBF neural network is introduced to online estimate the pseudo-partial derivative of the system data model. In order to better deal with the time lag problem in the infrared heating process, the RBF-MFAPC algorithm is proposed in combination with the rolling optimization mechanism of model predictive control (MPC). The rolling optimization prediction mechanism of model predictive control is used to process the time lag characteristics in the infrared heating process, and the design of the RBF-MFAPC controller is completed. Finally, the real-time collected temperature signal is used as the input of the RBF-MFAPC controller, and the infrared heater power is dynamically adjusted through real-time feedback, thereby achieving precise control of the temperature of the infrared heating process. This application does not rely on the model parameters of the infrared heating process. Through the nonlinear approximation of the system dynamic characteristics of the RBF neural network and the rolling optimization mechanism of the model predictive control, it significantly improves the control accuracy of the MFAC, achieves high-precision temperature control during the infrared heating process, and improves the adaptability of the system.

[0121] It should be noted that, for the purpose of simple description, the method embodiments are expressed as a series of action combinations, but those skilled in the art should know that the embodiments of the present application are not limited to the described order of actions, because according to the embodiments of the present application, certain steps can be performed in other orders or simultaneously.

[0122] like Figure 4 As shown, the embodiment of the present application further provides a model-free adaptive temperature control system 400 for an infrared heating process, comprising:

[0123] a conversion module 410 for converting the infrared heating system into a nonlinear discrete-time system according to input and output data of the infrared heating system, and converting the nonlinear discrete-time system into a dynamic linear data model;

[0124] a pseudo partial derivative estimation module 420 for online estimating pseudo partial derivatives in the dynamic linear data model using an RBF neural network to obtain estimated pseudo partial derivative values, wherein the pseudo partial derivative values ​​are output parameters obtained by the RBF neural network using nonlinear approximation capabilities to estimate the dynamic characteristics of the infrared heating system;

[0125] A predictive controller construction module 430 is configured to process the time lag characteristics of the infrared heating system during the infrared heating process using a model predictive control based on MFAC, and to construct a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data;

[0126] The analysis and prediction module 440 is used to take the heating data of the infrared heating system collected in real time as input, analyze and predict the behavior of the infrared heating system within a certain time period through the prediction controller, optimize the control input, and output dynamic adjustment information; wherein, the dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

[0127] Optionally, the conversion module is specifically used to: take the input and output data of the selected infrared heating system as a reference, according to y(k+1)=f(y(k),y(k-1),...y(kn y );u(k),u(k-1),...u(kn u )), convert the infrared heating system into a nonlinear discrete-time system; according to y(k+1)=y(k)+φ(k)Δu(k), convert the nonlinear discrete system of the infrared heating process into a compact dynamic linear data model; wherein k is the sampling time, y(k+1) is the actual output temperature of the infrared heating system at the next moment, f is an unknown nonlinear function, y(k) is the output temperature of the infrared heating system at the current moment, u(k) is the input power, y(k-1) is the output temperature of the infrared heating system at the previous moment, u(k-1) is the input power of the infrared heating system at the next moment, and n y Indicates the order of the system output, n u represents the order of the system input, φ(k) is the pseudo partial derivative of the system, Δu(k)=u(k)-u(k-1) and Δu(k)≠0.

[0128] Optional, pseudo partial derivative estimation module, including:

[0129] The pseudo partial derivative estimation submodule is used to use the information vector of the input and output data of the dynamic linear data model as the input of the RBF neural network, use the Gaussian basis function to determine the output of the hidden layer neurons through nonlinear mapping, and combine the connection weights between the neurons to linearly superimpose the hidden layer output to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative estimated at the current moment.

[0130] Optionally, the pseudo partial derivative estimation submodule is specifically used to:

[0131] x(k)=[y(k),…,y(km);u(k-1),…,u(kn)], construct the input and output data information vector as the RBF neural network input x(k); use the Gaussian basis function as the activation function, according to The input x(k) is nonlinearly mapped and The output of the hidden layer neurons is linearly superimposed to obtain the output layer output φ(k) of the RBF neural network as the estimated pseudo partial derivative value; where φ(k) is the estimated pseudo partial derivative value in the data model, m and n are both positive integers, representing the number of input and output data, c j represents the center vector of the jth hidden layer neuron; b j Indicates the radius of the jth hidden layer neuron, ω j (k) represents the connection weight between the jth neuron in the hidden layer and the neuron in the output layer at the current moment.

[0132] Optionally, a predictive controller construction module is specifically used to take the pseudo partial derivative and the real-time collected temperature data as input, according to Define a model-free adaptive predictive controller; according to Define the N-step forward prediction output equation of the infrared heating system; where N represents the output prediction step length of the system, represents the predicted output matrix of the system, represents the expected output matrix of the system, N u represents the control time domain constant of the system; Represents the predicted input change matrix of the system, E(k) represents the identity matrix, and A(k) represents the pseudo partial derivative prediction matrix.

[0133] Optional analysis and prediction module, including:

[0134] a pseudo partial derivative prediction value update submodule, configured to model the temporal variation pattern of the pseudo partial derivative value using a pseudo partial derivative prediction algorithm, update the pseudo partial derivative prediction value in combination with prediction error feedback correction, and update the prediction matrix in the prediction controller using the pseudo partial derivative prediction value;

[0135] The predictive control strategy updating submodule is used to substitute the dynamic linear data model, the updated pseudo partial derivative prediction matrix and the real-time collected heating data into the control input criterion function in the predictive controller, construct the control law expression, and use the first control quantity in the predicted input change matrix to update the predictive control strategy and output dynamic adjustment information.

[0136] Optional, pseudo partial derivative prediction value update submodule, specifically used to: Define the autoregressive model parameter matrix θ(k) and the autoregressive model pseudo-partial derivative matrix Φ(k-1); according to Get the autoregressive model parameter θ; take the autoregressive model parameter θ as input, according to

[0137] φ(k+j)=θ1(k)φ(k+j-1)+θ2(k)φ(k+j-2)+…+θ p (k)φ(k+jp) constructs a pseudo partial derivative prediction algorithm to obtain the pseudo partial derivative prediction value φ(k+j); where δ is a positive number and δ∈(0,1], φ(k+j) represents the pseudo partial derivative prediction value at time j, p represents the order of the autoregressive model, and j=1.

[0138] Optional, predictive control strategy update submodule, specifically used to: Construct control input criterion function J u ; Substitute the dynamic linear data model and the real-time collected heating data into the control input criterion function J u , and predict the input change matrix U Nu (k) Derivative, construct the control law expression by For input, according to Predict the first control variable in the input change matrix and obtain the control input u(k) at the current moment as dynamic adjustment information; where λ is the penalty factor of the system input, which is used to limit the variable of the control input and ensure the smoothness of the control input.

[0139]

[0140] Optionally, the model-free adaptive temperature control system 400 further includes:

[0141] The connection weight update module is used to use the one-step-ahead square error as the performance indicator function J, according to Determine the expected output signal of the system at the next moment; minimize the performance index function J, according to Update the connection weights, and the updated connection weights are used to update the pseudo partial derivative values; where y(k+1) is the actual output signal of the system at the next moment, and the connection weights ω from the hidden layer to the output layer are corrected by the gradient descent method. j,ω j (k+1) represents the connection weight at the next moment; α is the learning efficiency, and α∈(0,1], β is the inertia coefficient, and β∈(0,1], ω j (k-1) represents the connection weight at the previous moment, Δω j (k)=ω j (k)-ω j (k-1).

[0142] It should be noted that the model-free adaptive temperature control system for the infrared heating process provided in the embodiment of the present application can execute the model-free adaptive temperature control method for the infrared heating process provided in any embodiment of the present application, and has the corresponding functions and beneficial effects of the execution method.

[0143] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0144] The foregoing is merely a list of specific embodiments of the present application, intended to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the broadest scope consistent with the principles and novel features of the present application.

Claims

1. A model-free adaptive temperature control method for an infrared heating process, characterized in that: include: According to input and output data of the infrared heating system, the infrared heating system is converted into a nonlinear discrete time system, and the nonlinear discrete time system is converted into a dynamic linear data model; Using an RBF neural network to perform online estimation of pseudo partial derivatives in the dynamic linear data model to obtain estimated pseudo partial derivative values, wherein the pseudo partial derivative values ​​are output parameters obtained by the RBF neural network using nonlinear approximation capabilities to estimate the dynamic characteristics of the infrared heating system; Using MFAC-based model predictive control to process the time lag characteristics of the infrared heating system during the infrared heating process, and constructing a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data; Taking the real-time collected heating data of the infrared heating system as input, the predictive controller analyzes and predicts the behavior of the infrared heating system within a certain period of time, optimizes the control input, and outputs dynamic adjustment information; The dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

2. The method according to claim 1, characterized in that According to input and output data of the infrared heating system, the infrared heating system is converted into a nonlinear discrete time system, and the nonlinear discrete time system is converted into a dynamic linear data model, including: Based on the input and output data of the selected infrared heating system, according to y(k+1)=f(y(k),y(k-1),...y(kn y );u(k),u(k-1),...u(kn u )), converting the infrared heating system into a nonlinear discrete-time system; According to y(k+1)=y(k)+φ(k)Δu(k), the nonlinear discrete system of infrared heating process is converted into a compact dynamic linear data model; Where k is the sampling time, y(k+1) is the actual output temperature of the infrared heating system at the next moment, f is an unknown nonlinear function, y(k) is the output temperature of the infrared heating system at the current moment, u(k) is the input power, y(k-1) is the output temperature of the infrared heating system at the previous moment, u(k-1) is the input power of the infrared heating system at the next moment, and n y Indicates the order of the system output, n u represents the order of the system input, φ(k) is the pseudo partial derivative of the system, Δu(k)=u(k)-u(k-1) and Δu(k)≠0.

3. The method according to claim 1, characterized in that The RBF neural network is used to perform online estimation of the pseudo partial derivative in the dynamic linear data model to obtain estimated pseudo partial derivative values, including: The information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between neurons to obtain the output value of the RBF neural network at the current moment, which is used as the pseudo partial derivative value estimated at the current moment.

4. The method according to claim 3, characterized in that The information vector of the input and output data of the dynamic linear data model is used as the input of the RBF neural network, the output of the hidden layer neurons is determined by nonlinear mapping using the Gaussian basis function, and the hidden layer output is linearly superimposed in combination with the connection weights between the neurons to obtain the output value of the RBF neural network at the current moment as the pseudo partial derivative value estimated at the current moment, including: According to x(k)=[y(k),…,y(km);u(k-1),…,u(kn)], construct the input and output data information vector as the RBF neural network input x(k); The Gaussian basis function is used as the activation function, according to The input x(k) is nonlinearly mapped and The output of the hidden layer neurons is linearly superimposed to obtain the output layer output φ(k) of the RBF neural network as the estimated pseudo partial derivative; Among them, φ(k) is the estimated pseudo partial derivative in the data model, m and n are both positive integers, representing the number of input and output data, c j represents the center vector of the jth hidden layer neuron; b j Indicates the radius of the jth hidden layer neuron, ω j (k) represents the connection weight between the jth neuron in the hidden layer and the neuron in the output layer at the current moment.

5. The method according to claim 1, wherein The time lag characteristics of the infrared heating system during the infrared heating process are processed by using the MFAC-based model predictive control. A model-free adaptive predictive controller is constructed based on the pseudo partial derivative and the input and output data, including: Taking the pseudo partial derivative and the real-time collected temperature data as input, according to Define a model-free adaptive predictive controller; according to Define the N-step forward prediction output equation of the infrared heating system; Where N represents the output prediction step size of the system, represents the predicted output matrix of the system, represents the expected output matrix of the system, N u represents the control time domain constant of the system; Represents the predicted input change matrix of the system, E(k) represents the identity matrix, and A(k) represents the pseudo partial derivative prediction matrix.

6. The method according to claim 1, wherein Taking the real-time collected heating data of the infrared heating system as input, the predictive controller analyzes and predicts the behavior of the infrared heating system within a certain period of time, optimizes the control input, and outputs dynamic adjustment information, including: Modeling the time series variation law of the pseudo partial derivative value using a pseudo partial derivative prediction algorithm, updating the pseudo partial derivative prediction value in combination with prediction error feedback correction, and updating the prediction matrix in the prediction controller using the pseudo partial derivative prediction value; The dynamic linear data model, the updated pseudo partial derivative prediction matrix and the real-time collected heating data are substituted into the control input criterion function in the predictive controller to construct a control law expression, and the first control quantity in the predicted input change matrix is ​​used to update the predictive control strategy and output dynamic adjustment information.

7. The method according to claim 6, characterized in that The method includes: using a pseudo partial derivative prediction algorithm to model the time series variation law of the pseudo partial derivative value, updating the pseudo partial derivative prediction value in combination with prediction error feedback correction, and using the pseudo partial derivative prediction value to update the prediction matrix in the prediction controller, including: according to Define the autoregressive model parameter matrix θ(k) and the autoregressive model pseudo partial derivative matrix Φ(k-1); according to Get the autoregressive model parameters θ; Taking the autoregressive model parameter θ as input, according to φ(k+j)=θ1(k)φ(k+j-1)+θ2(k)φ(k+j-2)+…+θ p (k)φ(k+jp) constructs a pseudo partial derivative prediction algorithm and obtains the pseudo partial derivative prediction value φ(k+j); Where δ is a positive number and δ∈(0,1], φ(k+j) represents the pseudo partial derivative prediction value at time j, p represents the order of the autoregressive model, and j=1.

8. The method according to claim 6, characterized in that Substitute the dynamic linear data model, the updated pseudo partial derivative prediction matrix, and the real-time collected heating data into the control input criterion function in the predictive controller to construct a control law expression, and use the first control variable in the predicted input change matrix to update the predictive control strategy and output dynamic adjustment information, including: according to Construct control input criterion function J u ; Substitute the dynamic linear data model and the real-time collected heating data into the control input criterion function J u , and predict the input change matrix Derivative, construct control law expression by For input, according to Predict the first control variable in the input change matrix and obtain the control input u(k) at the current moment as dynamic adjustment information; Among them, λ is the penalty factor of the system input, which is used to limit the variable of the control input and ensure the smoothness of the control input.

9. The method according to claim 1, characterized in that After outputting the dynamic adjustment information, it also includes: The one-step-ahead square error is used as the performance indicator function J, according to Determine the expected output signal of the system at the next moment; Minimize the performance index function J, according to Updating the connection weights, wherein the updated connection weights are used to update the pseudo partial derivative values; Among them, y(k+1) is the actual output signal of the system at the next moment, and the connection weight ω from the hidden layer to the output layer is corrected by the gradient descent method. j ,ω j (k+1) represents the connection weight at the next moment; α is the learning efficiency, and α∈(0,1], β is the inertia coefficient, and β∈(0,1], ω j (k-1) represents the connection weight at the previous moment, Δω j (k)=ω j (k)-ω j (k-1).

10. A model-free adaptive temperature control system for an infrared heating process, characterized in that: include: a conversion module, configured to convert the infrared heating system into a nonlinear discrete time system according to input and output data of the infrared heating system, and to convert the nonlinear discrete time system into a dynamic linear data model; a pseudo partial derivative estimation module, configured to perform online estimation of the pseudo partial derivatives in the dynamic linear data model using an RBF neural network to obtain estimated pseudo partial derivative values, wherein the pseudo partial derivative values ​​are output parameters obtained by the RBF neural network using a nonlinear approximation capability to estimate the dynamic characteristics of the infrared heating system; A predictive controller construction module is used to process the time lag characteristics of the infrared heating system during the infrared heating process by using the MFAC-based model predictive control, and to construct a model-free adaptive predictive controller based on the pseudo partial derivative and the input and output data; an analysis and prediction module, configured to use the real-time collected heating data of the infrared heating system as input, analyze and predict the behavior of the infrared heating system within a certain time period through the prediction controller, optimize the control input, and output dynamic adjustment information; The dynamic adjustment information is used as a control strategy to provide real-time feedback and dynamically adjust the power of the infrared heater in the infrared heating system, and the heating data at least includes temperature data and power data during the infrared heating process.

Images